BioBox MCP: Institutional Memory for Agents

Complex portfolio evaluation decisions don’t occur in a single agentic session. They are the accumulation of decisions made by cross functional teams against frameworks that are constantly evolving and multi-modal data that is growing over time. Skills files struggle to perform under the dynamic nature of portfolio evaluation. We report these findings in this white paper.

Key Findings

A skills file lets an agent approximate your scientific framework. The MCP lets it execute your framework on your data foundation, and track which version ran against the data and knowledge available at the time of the decision. When it comes to complex portfolio evaluation decisions, approximation is where the risk lives. The MCP setup passed 19 of 19 evaluation assertions and the skills setup passed 4. Both handled direct lookups identically, so the gap sits entirely in the harder work: composing evidence across modalities, applying a decision framework as the scientist intends, and reconstructing a decision months later.